Improving Speech Intelligibility Using Ideal Binary Mask

Abstract

Background: The application of the ideal binary mask (IBM) for speech signal processing provides remarkable intelligibility improvements in both normal-hearing and hearing-impaired listeners. Binary mask widely applies to the time-frequency (T–F) representation of a noisy signal and eliminates units of a signal below a signal-to-noise-ratio (SNR) threshold while retains others. Methods: The factors underlying intelligibility of ideal binary-masked speech were examined and evaluated in the present study. The effects of the local SNR threshold, input SNR level, masker type, and ideal mask-estimator were examined. New estimators including weighted Euclidean and COSH were proposed in which, the human perceptual auditory masking effect and perceptual perception were incorporated. Findings: High-performance plateau for SNR thresholds ranging from −20 to 5 dB was observed. Findings could be used for hearing-aid and cochlear-implant designs. Conclusion: Intelligibility of speech was high even at −10 dB SNR for all maskers tested. Performance assessment shows that our proposed estimators can achieve more significant noise estimation as compared to the Wiener estimator.

Keywords

• Speech enhancement
• Binary masking
• Speech intelligibility